Merge pull request #1108 from olamarre/bugfix/ep-pickling

fix: pickle and deep copy classification models with EP

CHANGELOG.md

@@ -1,6 +1,8 @@
 # Changelog
 
 ## Unreleased
+* fix pickle and deep copy for classification models inheriting from EP #1108 [olamarre]
+
 * update prior `__new__` methods #1098 [MartinBubel]
 
 * fix invalid escape sequence #1011 [janmayer]

GPy/inference/latent_function_inference/expectation_propagation.py

@@ -229,24 +229,17 @@ def _stop_criteria(self, ga_approx):
         v_diff = np.mean(np.square(ga_approx.v-self.ga_approx_old.v))
         return ((tau_diff < self.epsilon) and (v_diff < self.epsilon))
 
-    def __setstate__(self, state):
-        super(EPBase, self).__setstate__(state[0])
-        self.epsilon, self.eta, self.delta = state[1]
-        self.reset()
-
-    def __getstate__(self):
-        return [super(EPBase, self).__getstate__() , [self.epsilon, self.eta, self.delta]]
-
     def _save_to_input_dict(self):
-        input_dict = super(EPBase, self)._save_to_input_dict()
-        input_dict["epsilon"]=self.epsilon
-        input_dict["eta"]=self.eta
-        input_dict["delta"]=self.delta
-        input_dict["always_reset"]=self.always_reset
-        input_dict["max_iters"]=self.max_iters
-        input_dict["ep_mode"]=self.ep_mode
-        input_dict["parallel_updates"]=self.parallel_updates
-        input_dict["loading"]=True
+        input_dict = {
+            "epsilon": self.epsilon,
+            "eta": self.eta,
+            "delta": self.delta,
+            "always_reset": self.always_reset,
+            "max_iters": self.max_iters,
+            "ep_mode": self.ep_mode,
+            "parallel_updates": self.parallel_updates,
+            "loading": True
+        }
         return input_dict
 
 class EP(EPBase, ExactGaussianInference):

GPy/testing/test_inference.py

@@ -5,6 +5,8 @@
 The test cases for various inference algorithms
 """
 
+import copy
+import pickle
 import numpy as np
 import GPy
 
@@ -146,6 +148,32 @@ def test_inference_EP(self):
             < 1e6
         )
 
+    def test_pickle_copy_EP(self):
+        """Pickling and deep-copying a classification model employing EP"""
+
+        # Dummy binary classification dataset
+        X = np.array([0, 1, 2, 3]).reshape(-1, 1)
+        Y = np.array([0, 0, 1, 1]).reshape(-1, 1)
+
+        # Some classification model
+        inf = GPy.inference.latent_function_inference.expectation_propagation.EP(
+            max_iters=30, delta=0.5
+        )
+        m = GPy.core.GP(
+            X=X,
+            Y=Y,
+            kernel=GPy.kern.RBF(input_dim=1, variance=1.0, lengthscale=1.0),
+            inference_method = inf,
+            likelihood=GPy.likelihoods.Bernoulli(),
+            mean_function=None
+        )
+        m.optimize()
+
+        m_pickled = pickle.dumps(m)
+        assert pickle.loads(m_pickled) is not None
+
+        assert copy.deepcopy(m) is not None
+
     # NOTE: adding a test like above for parameterized likelihood- the above test is
     # only for probit likelihood which does not have any tunable hyperparameter which is why
     # the term in dictionary of gradients: dL_dthetaL will always be zero. So here we repeat tests for